1 / 59

Index Construction

Index Construction. Adapted from Lectures by Prabhakar Raghavan (Google) and Christopher Manning (Stanford). Plan. Last lectures: Dictionary data structures Tolerant retrieval Wildcards Spell correction Soundex This time: Index construction. n-z. a-hu. hy-m. $m. mace. madden.

aira
Download Presentation

Index Construction

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Index Construction Adapted from Lectures by Prabhakar Raghavan (Google) and Christopher Manning (Stanford) L06IndexConstruction

  2. Plan • Last lectures: • Dictionary data structures • Tolerant retrieval • Wildcards • Spell correction • Soundex • This time: • Index construction n-z a-hu hy-m $m mace madden mo among amortize on abandon among L06IndexConstruction

  3. Index construction • How do we construct an index? • What strategies can we use with limited main memory? • Hardware Basics • Many design decisions in information retrieval are based on the characteristics of hardware • We begin by reviewing hardware basics L06IndexConstruction

  4. Hardware basics • Access to data in memory is much faster than access to data on disk. (prefer caching) • Disk seeks: No data is transferred from disk while the disk head is being positioned. • Therefore: Transferring one large chunk of data from disk to memory is faster than transferring many small chunks. • Disk I/O is block-based: Reading and writing of entire blocks (as opposed to smaller chunks). • Block sizes: 8KB to 256 KB. L06IndexConstruction

  5. Hard disk geometry and terminology L06IndexConstruction

  6. Hardware basics • Servers used in IR systems now typically have several GB of main memory, sometimes tens of GB. • Available disk space is several (2–3)orders of magnitude larger. • Fault tolerance is very expensive: It’s much cheaper to use many regular machines rather than one fault tolerant machine. L06IndexConstruction

  7. Hardware assumptions • symbol statistic value • s average seek time 5 ms = 5 x 10−3 s • b transfer time per byte 0.02 μs = 2 x 10−8 s • processor’s clock rate 109 s−1 • p low-level operation 0.01 μs = 10−8 s (e.g., compare & swap a word) • size of main memory several GB • size of disk space 1 TB or more • Mem. trans. time per byte 5 ns L06IndexConstruction

  8. RCV1: Our corpus for this lecture • Shakespeare’s collected works definitely aren’t large enough. • The corpus we’ll use isn’t really large enough either, but it’s publicly available and is at least a more plausible example. • As an example for applying scalable index construction algorithms, we will use the Reuters RCV1 collection (Approx. 1GB). • This is one year of Reuters newswire (part of 1996 and 1997) L06IndexConstruction

  9. A Reuters RCV1 document L06IndexConstruction

  10. Type/token distinction • Token– an instance of a word occurring in a document • Type – an equivalence class of tokens • In June, the dog likes to chase the cat in the barn. • 12 word tokens, 8 word types • Tokens– {the, in} = Types L06IndexConstruction

  11. Reuters RCV1 statistics • symbol statistic value • N documents 800,000 • L avg. # tokens per doc 200 • M terms (= word types) 400,000 • avg. # bytes per token 6 (incl. spaces/punct.) • avg. # bytes per token 4.5 (without spaces/punct.) • avg. # bytes per term 7.5 • non-positional postings = term occurrences 100M 4.5 bytes per word token vs. 7.5 bytes per word type: why?

  12. Recall IIR1 index construction • Documents are parsed to extract words and these are saved with the Document ID. Doc 1 Doc 2 I did enact Julius Caesar I was killed i' the Capitol; Brutus killed me. So let it be with Caesar. The noble Brutus hath told you Caesar was ambitious L06IndexConstruction

  13. After all documents have been parsed, the inverted file is sorted by terms. Key step We focus on this sort step. We have 100M items to sort. L06IndexConstruction

  14. Scaling index construction • In-memory index construction does not scale. • How can we construct an index for very large collections? • Taking into account the hardware constraints we just learned about . . . • Memory, disk, speed etc. L06IndexConstruction

  15. Use the same algorithm for disk? • Can we use the same index construction algorithm (internal sorting algorithms) for larger collections, but by using disk instead of memory? • No: Sorting T = 100,000,000 records on disk is too slow – too many disk seeks. • We need an external sorting algorithm. If every comparison took 2 disk seeks, and N items could be sorted with N log2N comparisons, how long would this take? L06IndexConstruction

  16. Disk Sorting Time Estimate • Approx. 2 * 5 milliseconds * 100M * log2 100M = 2 * 5 milliseconds * 100M * 27 as log2 10 = 3.32 (log10 2 = 0.301) = 27000000 seconds = 7500 hours (= 10 months) L06IndexConstruction

  17. BSBI: Blocked sort-based Indexing (Sorting with fewer disk seeks) • Assume that the entire dictionary can fit into the memory • Parse documents into (termId, docId) pairs to conserve space, maintaining (term -> termId) universal mapping in the main memory. • Eventually 12-byte (4+4+4) records of the form (termId, doc, freq) are generated as we parse docs. • term requires more space than termId

  18. (cont’d) • As we build the index, we parse docs one at a time. • While building the index, we cannot easily exploit compression tricks (you can, but much more complex) • The final postings for any term are incomplete until the end. • At 12 bytes per postings entry, demands a lot of space for large collections. • In general, need to store intermediate results on disk. • However, T = 100,000,000 in the case of RCV1 • So … we can do this in memory in 2008, but typical collections are much larger. E.g. New York Times provides index of >150 years of newswire

  19. BSBI: Blocked sort-based Indexing (Sorting with fewer disk seeks) • Must now sort 100M such 12-byte records by term. • Define a Block ~ 10M of such records • Can easily fit a couple into memory. • Will have 10 such blocks to start with. • Basic idea of algorithm: • Accumulate postings for each block, sort, write to disk. • Then merge the blocks into one long sorted order.

  20. sortchunks mergechunks split On-disk sorting • Break up list into chunks. Sort chunks, then merge chunks • We can pick the chunk size so that we can sort it in memory

  21. L06IndexConstruction

  22. L06IndexConstruction

  23. 1 3 2 4 How to merge the sorted runs? • Can do binary merges, with a merge tree of height log2n (e.g., log210 = 4 layers). • During each layer, read into memory runs in blocks of 10M, merge, write back. 2 1 Merged run. 3 4 Runs being merged. Disk

  24. How to merge the sorted runs? • But it is more efficient to do a n-way merge, where you are reading from all blocks simultaneously • Providing you read decent-sized chunks of each block into memory, you’re not killed by disk seeks Binary Merge N-way Merge L06IndexConstruction

  25. Remaining problem with sort-based algorithm • Our assumption was: we can keep the dictionary in memory. • We need the dictionary (which grows dynamically) to map a term to termID. • Actually, we could work with term,docID postings instead of termID,docID postings . . . • . . . but then intermediate files become very large. (We would end up with a scalable, but very slow index construction method.) L06IndexConstruction

  26. SPIMI: Single-pass in-memory indexing • Key idea 1: Generate separate dictionaries for each block – no need to maintain term-termID mapping across blocks. • Key idea 2: Don’t sort. Accumulate postings in postings lists as they occur. • With these two ideas we can generate a complete inverted index for each block. • These separate indexes can then be merged into one big index. L06IndexConstruction

  27. SPIMI-Invert • Merging of blocks is analogous to BSBI but what about sorted postings?

  28. SPIMI: Compression • Compression makes SPIMI even more efficient. • Compression of terms • Compression of postings • Requires sorted postings? • See next lecture L06IndexConstruction

  29. Distributed indexing • For web-scale indexing: must use a distributed computing cluster • Individual machines (stock hardware) are fault-prone • Can unpredictably slow down or fail • How do we exploit such a pool of machines? • ‘Embarrassing parallelism’ in the data to rescue L06IndexConstruction

  30. Google data centers • Google data centers mainly contain commodity machines and are distributed around the world. • Estimate: a total of 1 million servers, 3 million processors/cores (Gartner 2007) • Estimate: Google installs 100,000 servers each quarter. • Based on expenditures of 200–250 million dollars per year • This would be 10% of the computing capacity of the world!?! L06IndexConstruction

  31. Google data centers • If in a non-fault-tolerant system with 1000 nodes, each node has 99.9% uptime, what is the uptime of the system? • Answer: 36% • Note, (1 – 0.9991000) = 0.63 • Probability that at least one will fail. L06IndexConstruction

  32. Distributed indexing • Maintain a master machine directing the indexing job – considered “safe”. • Break up indexing into sets of (parallel) tasks. • Master machine assigns each task to an idle machine from a pool. • Also manages faults and scheduling redundant data / computations automatically L06IndexConstruction

  33. Parallel tasks • We will use two sets of parallel tasks • Parsers • Inverters • Break the input document corpus into splits • Each split is a subset of documents (corresponding to blocks in BSBI/SPIMI) L06IndexConstruction

  34. Parsers • Master assigns a split to an idle parser machine • Parser reads a document at a time and emits (term, doc) pairs • Parser writes pairs into j partitions • Each partition is for a range of terms’ first letters • (e.g., a-f, g-p, q-z) – here j=3. • Now to complete the index inversion L06IndexConstruction

  35. Inverters • An inverter collects all (term,doc) pairs (= postings) for one term-partition. • Sorts and writes to postings lists L06IndexConstruction

  36. Data flow Master assign assign Postings Parser Inverter a-f g-p q-z a-f Parser a-f g-p q-z Inverter g-p Inverter splits q-z Parser a-f g-p q-z Map phase Reduce phase Segment files

  37. MapReduce • The index construction algorithm we just described is an instance of MapReduce. • MapReduce (Dean and Ghemawat 2004) is a robust and conceptually simple framework for • distributed computing … • … without having to write code for the distribution part. • They describe the Google indexing system (ca. 2002) as consisting of a number of phases, each implemented in MapReduce. L06IndexConstruction

  38. Schema for index construction in MapReduce • Schema of map and reduce functions • map: input → list(k, v) reduce: (k,list(v)) → output • Instantiation of the schema for index construction • map: web collection → list(termID, docID) • reduce: (<termID1, list(docID)>, <termID2, list(docID)>, …) → (postings list1, postings list2, …) • Example for index construction • map: d2 : C died. d1 : C came, C c’ed. → (<C, d2>, <died,d2>, <C,d1>, <came,d1>, <C,d1>, <c’ed, d1> • reduce: (<C,(d2,d1,d1)>, <died,(d2)>, <came,(d1)>, <c’ed,(d1)>) → (<C,(d1:2,d2:1)>, <died,(d2:1)>, <came,(d1:1)>, <c’ed,(d1:1)>) L06IndexConstruction

  39. Dynamic indexing • Up to now, we have assumed that collections are static. • They rarely are: • Documents come in over time and need to be inserted. • Documents are deleted and modified. • This means that the dictionary and postings lists have to be modified: • Postings updates for terms already in dictionary • New terms added to dictionary L06IndexConstruction

  40. Simplest approach • Maintain “big” main index • New docs go into “small” auxiliary index • Search across both, merge results • Deletions • Invalidation bit-vector for deleted docs • Filter docs output on a search result by this invalidation bit-vector • Periodically, re-index into one main index L06IndexConstruction

  41. Merging Aux: Main:

  42. Merging Aux: Main:

  43. Merging Aux: Main:

  44. Merging Aux: Main: Every time we merge we use the entire index Can we do better?

  45. Logarithmic merge • Maintain a series of indexes, each twice as large as the previous one. • Keep smallest (Z0) in memory • Larger ones (I0, I1, …) on disk • If Z0 gets too big (> n), write to disk as I0 • or merge with I0 (if I0 already exists) as Z1 • Either write merged Z1 to disk as I1 (if no I1) • Or merge with I1 to form Z2 • etc. L06IndexConstruction

  46. Logarithmic merge

  47. Logarithmic merge main memory

  48. Logarithmic merge main memory 1 2 3 4 5 …

  49. Logarithmic merge main memory 1 2 3 4 5 …

  50. Logarithmic merge main memory 1 2 3 4 5 …

More Related